Interstitial lung disease associated with connective tissue diseases: The use of statistical structure analysis in model development

Interstitial lung disorders associated with the connective tissue diseases are thought to be quite common, but their precise analysis is fraught with difficulty because of the absence of a Gold Standard that, short of open lung biopsy, is not available. Analysis is hampered by the biologic variability in test results, large overlap between the normal and the disease population for individual tests, the confounding effect of smoking, and the complexities of viewing multidimensional data. In order to better define the pattern of the lung involvement, an entirely different approach has been adopted that is based on the use of graphic and clustering techniques to define the multivariate structure inherent in the data, then discriminant analysis to assign patients into distinct clusters. Using this approach it has been possible to group patients into four distinct clusters based on the results of respiratory function studies, gallium lung scan, bronchoalveolar lavage, and smoking status. These clusters are a normal smoking cluster and a normal nonsmoking cluster, a cluster of patients with active ILD, and a cluster with bronchiolitis. The validity of this method has been verified, and an algorithm has been developed that allows the assignment of any new patient entering the study into one of these clusters. This type of analysis offers a valuable new approach to the categorization of patients into distinct groups based on the results of multiple investigations.